Following Staffan Lindeberg's posting of  Dietary Links to
Alzheimer's Disease by William B. Grant, PhD, I scanned the article,
focusing on its statistical methods and results.  My apologies if
this is too technical, I do think the conclusions are worth
examining.

I hardly agree with the conclusions of this article, though I
accept the many linkages it develops between oxidative stress and
Alzheimer's disease (among those 65+).

Its results suggest that among the factors related to Alzheimer's
are total caloric intake and fat intake.

There are 11 data points, 7 in the European/North American cluster.
 The data are highly aggregate measures of diet and Alzheimer's
incidence showing national rates.  They are crudely adjusted for age
distribution.  There are few degrees of freedom because of the
small sample size.  Consequently, outliers are influential.  The
equation for fat is: AD prevalence rate = -0.203+(0.0312*fat
(grams/day)).  A large negative intercept term and a positive slope,
leading to the author's conclusion.  A large negative intercept
often indicates a non-linear relationship (which these data surely
exhibit).  Because fat intake varies so much among countries,
compared to total caloric intake, the differences among countries
are expanded in the regressions using fat as the dependent variable.
 This exaggerates the influence of outliers.

Take a look at this graph from the ariticle
http://www.coa.uky.edu/ADReview/Grant_Images/grantf1.gif and you
will see that there are two groups: developed Northern American and
European countries and Asian and African countries.  There are no
points in the middle and a vast gulf between the data clusters.

Within the developed cluster there is no relation between fat
consumption and Alzheimer's.

Measurement error is all over these data: different reporting
rates, more thorough surveys, and genetic composition of the
populations are just a few.

This is compounded by multicolinearity: fat and total calories are
highly correlated and so is the % of cereals.  You can't put them in
the same equation and get any precision in your estimates.  The
relationship between total calories and Alzheimer's is less
problematic (statistically speaking).  But, Europeans and Americans
are bigger and must eat more than Asians and Nigerian pastoralists.

Grant's article is excellent on other grounds, but the statistics
show nothing, or they hint at a genetic disposition to Alzheimer's
more than an environmental one.

Arthur De Vany
Professor
 <[log in to unmask]>
http://www.socsci.uci.edu/mbs/personnel/devany/devany.html
University of California
Institute for Mathematical Behavioral Sciences
3151 Social Science Plaza
Irvine, CA  92697-5100